Running this app on HPC environments without Docker

[jmanuelfuentes]

Hi,

Is there any way to run the app in an HPC environment without relying on Docker? Could alternatives like Singularity or Apptainer be used effectively in this context? Any recommendations or guidance would be greatly appreciated.

Thanks in advance!

[danifranco]

Hi,

For running BiaPy in an HPC cluster I recommend you to avoid using the GUI (this github project relates to the GUI). For running BiaPy in an HPC enviroment you have a few options. Here are my recommendations in order of what I would try first:

1. You can run BiaPy via command line as is described in our documentation. For example, here a link to instance segmentation workflow's how to run section in the doc. This is normally how I run BiaPy when I'm dealing with big datasets and need to use a cluster. Apart from that, this way of running BiaPy allows you to use more than one GPU, i.e. multiGPU (DDP), so if you want to use like that you would need to run it as described in the link too:

   

   It is true that the drawback of this option is that you need to have BiaPy's conda environment installed but you will have more flexibility.

2- If you really want to use a container you can convert the Docker container (BiaPy's Dockerfile here) into Singularity/Apptainer. Never did that but when I used Singularity years ago Docker containers could be converted into Singularity containers. I did a quick search and seems that it is still possible. Here is a tutorial I found from the University of Calgary.

Hope this helps you!